Moscow Polytechnic University has launched a pilot project titled Multi-Agent Design Bureau. The initiative involves teams of AI agents performing engineering tasks, while human engineers act as supervisors who set objectives, provide feedback, and refine calculations.
According to Anton Nalivayko, Vice-Rector for Research, the project aims to reduce the development cycle from the initial idea to a validated prototype from a range of 2–14 years to 1.5–4 months. This acceleration is achieved by having agents perform design, calculation, and technological tasks simultaneously rather than sequentially.
The AI agents operate in swarms, sharing project data and verifying the consistency of their solutions against initial requirements. For instance, an agent-technologist can identify manufacturing constraints, triggering automatic design adjustments and material re-evaluations to ensure structural integrity and manufacturability.
Currently, university engineers are training the agents using reference data, established calculation methods, and design tools. Each agent undergoes validation on test tasks before its skills are finalized. The entire workflow is conducted within the university's secure internal infrastructure, which stores all project documentation, decision logs, and agent versions.
The pilot program is scheduled to run until the end of 2026. Following its conclusion, the university plans to integrate the multi-agent design bureau across all its scientific and technical departments to automate routine operations and enable engineers to manage multiple complex projects simultaneously.
